Claude 3.7 Sonnet AI Tested On Pokémon Red

There are a lot of AI models doing the rounds, but Claude might be the only one tested on a Pokémon game. Specifically, Claude 3.7 Sonnet was tested on the classic Gameboy title, Pokémon Red, and its AI capabilities allowed it to progress and even battle several game bosses.

Claude 3.7 Sonnet is made by Anthropic, an AI safety and research company, and it’s the first hybrid reasoning model available for both personal and business use. There are three AI models in total, with the 3.7 version of Sonnet being the latest and also the only one so far to use what Anthropic calls “extended thinking.”

“Extended thinking mode isn’t an option that switches to a different model with a separate strategy. Instead, it’s allowing the very same model to give itself more time, and expend more effort, in coming to an answer,” the company said in a recent blog post. 

Gotta Catch ‘Em All, Claude

Extended thinking can be toggled on and off in Claude 3.7 Sonnet, but Anthropic wanted to see how much better it made the model by leaving it on and setting it up to play Pokémon Red. 

“We equipped Claude with basic memory, screen pixel input, and function calls to press buttons and navigate around the screen, allowing it to play Pokémon continuously beyond its usual context limits, sustaining gameplay through tens of thousands of interactions.”

The results were surprising, as previous versions of Claude couldn’t even figure out how to progress past the game’s starting area, whereas 3.7 Sonnet not only made its way out of the opening level of Pallet Town but successfully completed three other areas by defeating three Gym Leaders (game bosses).

What This Means For Future Use

The upshot of all this is that Claude 3.7 Sonnet’s extended thinking allows it to accomplish open-ended goals by maintaining focus. This also means that developers (of video games, as well as other industries) could use this AI for a variety of tasks, such as code generation and automatic language translation.